The mechanism of question answering service to support prebuilt question answering

Question

We checked this doc https://learn.microsoft.com/en-us/azure/cognitive-services/language-service/question-answering/overview, the question answering support two capabilities:

1. Custom question answering: Query answers from the populated question and answer pairs.
2. Prebuilt question answering: This capability allows users to get a response by querying a text passage without having the need to manage knowledgebases.

These two capabilities respond to these two populated data sources:

1.Import questions and answers
2. Import a project
Refer to https://learn.microsoft.com/en-us/azure/cognitive-services/language-service/question-answering/how-to/export-import-refresh

In our QA scenarios, we have some question and answer pairs, and also some releated docs. We want combine these two data sources into the answer candidates. We also found the question and answering leverage layered ranking: Azure search and NLP re-ranking model to give the potential answers.

This mechanism is straightforward for QA pairs. Because we could use these pairs to train the re-ranker model with listwise loss. BTW, which model we use here for re-ranker. But for releated docs, because we don't have pairs for re-ranker to train. How these docs will impact the final answers? Will these docs be fed into re-ranker to get the final ranking result. Or will these docs be extracted or parsed into QA pairs?

We are pretty curious about these mechanisms in order to better prepare the data fed to the model. Thanks!

Answer 1

@Wangxiang Ding Thanks for the question. You can add both the question and answer pairs, and also some related docs. Are you facing any error if yes please add more details.

As shown in the language studio support the below formats.